Predictions Of The Pathological Response To Neoadjuvant Chemotherapy For Breast Cancer Based On Complex Clinical Data

Breast cancer is a major malignant tumor affecting the health of women, neoadjuvant chemotherapy is a systemic and systematic drug treatment, commonly used before radiotherapy or local surgical treatment. According to the clinical research, Patients with pathologically complete response from neoadjuvant chemotherapy have a good prognosis. This shows that the pathological complete response in these patients may represent an alternative prognostic indicator. However, not all patients can obtain the pathological complete response from neoadjuvant chemotherapy. We need to conduct a research about the prediction of whether patients can obtain pathological complete response, thus to carry out the targeted subsequent treatment. According to the clinical pathological data of patients with breast cancer, using machine learning model to predict pathological response for neoadjuvant chemotherapy in the middle and later stages, may assist in the diagnosis of breast cancer, is helpful to improve the artificial diagnosis accuracy and efficiency, so as to improve the efficacy of tumor therapy. Therefore, we can use intelligent computing methods to build a prediction model with clinical data to divine the pathological response to Neoadjuvant Chemotherapy for breast cancer.The data of neoadjuvant chemotherapy for patients with breast cancer in The First Hospital of Jilin University were collected in this paper. The collected data contains a number of indicators. They are the basic situation of breast cancer patients, physical, ultrasound, mammography and pathology data before and after neoadjuvant chemotherapy, and pathology data after surgery. The collected data of neoadjuvant chemotherapy for breast cancer patients were processed by data integrity screening and data preprocessing. A total of 259 samples were composed of the experimental data set.Firstly the data set were divided according to roughly equal proportion for the established data sets, then we tried the K nearest neighbor (KNN), decision tree (C4.5), support vector machine (SVM) and random forest (RF) and other classification methods to establish the model. The classification accuracy of these models ranged from 60.94% to 68.75%, and the classification results were heavily inclined to PR class, did not achieve the purpose of classification.In order to solve the above problems, refer to the CR class which had the least number of samples, we divide data sets in an equal way,18 sub training sets are formed. After that we respectively use KNN, C4.5, SVM and RF to establish 18 single classification prediction model. Then two ensemble mechanisms, Adaboost and majority voting, have been established for the ensemble of various types of single model. The experimental results show that the classification accuracy of the integrated model was better than that of the single model. In addition, the classification accuracy of the majority voting ensemble model was better than that of the corresponding Adaboost ensemble model. The ensemble model (EKNN) based on the KNN using majority voting strategy achieved the best classification accuracy. The classification accuracy of the EKNN model for pathologic response after neoadjuvant chemotherapy for breast cancer is 77.78%, Kappa coefficient is 0.67. This model has good prediction ability and is capable to predict the pathological response of breast cancer after neoadjuvant chemotherapy.